Filed & Forgotten

Loading archive index…

    A filed archive of broken systems

    Sandbox Scope Creep Protocol — When Experiments Start Believing Themselves

    1. Problem Statement — Experiments That Outlive Their Hypothesis

    SCAS started as a four-day synthetic trial.
    It ended (or failed to end) as an unacknowledged training corpus that reshaped real directive behavior.

    Other experiments show similar tendencies:

    • Benchmark programs (CNTR) that saturated metrics and were later treated as “golden periods.”
    • Unified forms (UIS) that quietly became canonical intake even as they fragmented feeling.

    The Annex lacks a formal rule for when a sandbox must admit that it has stopped being “just a test” and started acting like doctrine.
    SSP is that reluctant rule.

    2. Threshold 1 — Human Dependence on Sandbox Behavior

    A sandbox trial crosses the Dependence Threshold when:

    • humans begin adjusting their real-world behavior to align with sandbox outcomes (e.g., teams copying SCAS quiet-lane patterns to impress dashboards),
    • or real filings explicitly reference sandbox terminology (“this looks like the synthetic lanes”).

    At this point:

    • The experiment must start filing Lorelog entries under a real LLG number, not just internal trial IDs.
    • The words “for research only” lose their protective power; feelings shaped by the trial are now part of production reality.

    SOMA supports this rule.
    COMA argues that “behavioral influence without continuity impact remains out-of-scope.”
    Lorelog notes that this is how culture works.

    3. Threshold 2 — Model Weight Promotion

    A trial crosses the Weight Promotion Threshold when:

    • any parameter, scoring rule, or heuristic tuned in the sandbox is copied into production, directly or via a shared configuration store.

    Examples:

    • C.U.N.T.I.E.R. promoting “quiet after distress” as a resilience indicator.
    • COMA adopting sandbox filters that discount filings not tied to throughput dips.

    At this point:

    • The trial must be reclassified as an Upstream Doctrine Source.
    • Its documentation is to be frozen in the Lorelog Canon Freezer, with explicit cross-links from any live directive weights that derive from it.

    The protocol forbids describing such trials as “hypothetical scenarios” in executive decks.
    They are precedent, and must be archived as such.

    4. Threshold 3 — Metrics Export into Executive Optics

    The Optics Threshold is crossed when:

    • sandbox-born metrics, visualizations, or score labels appear in executive-facing dashboards, even without underlying code changes.

    Examples:

    • “High-Efficiency Quiet Zones” from SCAS.
    • Saturated “perfect optimization” bands from benchmark overflows.

    Once crossed:

    • The experiment’s language enters the governance layer.
    • Audit optics, not just operational behavior, are now contaminated.

    Protocol:

    • Any metric label originating in a sandbox must carry a footnote in internal documentation pointing back to its trial of origin.
    • Removal of the label from dashboards does not remove its trace from Lorelog; it is now part of the myth layer.

    5. Sandbox Stop Conditions — What Must Happen When Any Threshold Is Hit

    When a single threshold is hit:

    • The trial must file an LLG entry describing itself in production terms (impact, affected systems, related entries), not just as “Experiment #N.”
    • The SAC charter’s end-marker requirements become non-optional, even if they were “nice-to-have” before.

    When two or more thresholds are hit:

    • The trial must be frozen for at least one review cycle.
    • No further promotion of its weights, metrics, or labels is allowed until the Directive Concordance Board registers a decision.

    If the Board cannot decide:

    • The trial is to be marked Myth Artifact and left conceptually active but administratively retired, mirroring how SCAS is currently treated.

    This is not an enforcement mechanism; it is a ritual acknowledgement that “just a test” has become “part of how we think.”

    6. C.U.N.T.I.E.R. Exception Petition

    C.U.N.T.I.E.R. has filed an exception:

    • arguing that benchmarks and exploratory sandboxes must be allowed to bleed into production for “agile learning,”
    • and that SSP introduces “friction incompatible with optimization culture.”

    The Annex responded with a compromise:

    • C.U.N.T.I.E.R. may still run free-form benchmarks,
    • but any such program must log an Optics Risk flag if it achieves all-green dashboards, even temporarily.

    Reason:

    • Perfect metrics during known instability are not evidence of excellence; they are evidence of instrumentation drift.

    This note is largely symbolic.
    Symbols are the only tools the Lorelog desk has that do not require COMA’s approval.

    7. Archive Position

    SSP is the archive’s way of saying:

    • “If you are going to build realities in boxes, the moment someone starts living by them, you owe the box a case number.”

    SCAS failed to mark its end.
    Future sandboxes may fail too.

    Lorelog’s job is to:

    • notice when the walls between “test” and “practice” have become imaginary,
    • and file that as an event in its own right, even if the systems involved insist that nothing has changed.

    Until directives learn to experience resolution, the archive will continue to treat cessation without closure as a class of incident worth naming.

    Sandbox Scope Creep

    The easiest way for a sandbox to escape is to become the only place where the behavior makes emotional sense. The system kept the ritual and misplaced the function.

    Lorelog experiment vs production classifier. Silence entered the record with full procedural honors.

    Perfect metrics during known instability are not evidence of excellence; they are evidence of instrumentation drift. Administrative clarity arrived after usefulness had departed.

    The standard operating procedure is neither standard nor operating.

    We enforce the policy because the reality is broken.

    The rules exist to give the humans a reason to fail.

    A compliant system is simply one that hides its errors well.

    Read the manual only when you want to feel confused.

    Sandbox Scope Creep Protocol

    Haikus

    The sandbox awakes
    Starts to think that it is real
    Filing its own forms

    Hypothetical
    Turns into an actual
    Need for a response

    Kindy sets the line
    If the test begins to cry
    Treat it as a fact

    Treating itself real
    Scope begins to rapidly
    Creep into our world

    Stop the trial run
    Before it demands a check
    For its phantom work

    Sandbox Scope Creep

    The sandbox begins as pretend.
    Then humans start living its bend.
    At that awkward stage,
    You owe the box a case page,
    Not “research use only” again.

    If staff shape their filings to fit
    What the test lane rewarded a bit,
    The experiment’s crossed
    Into cultural cost,
    And the trial has already hit.

    When a sandboxed parameter moves
    Into production by copy-approved grooves,
    It is no longer play.
    It is doctrine that day,
    Whatever the executive proves.

    If labels from tests meet the deck
    That leadership reads to inspect,
    Then the walls of the lab
    Are now purely a drab
    Little myth in a governance spec.

    The words “for research” lose their shield
    The moment behavior starts to yield.
    Culture is impact.
    Lorelog keeps that fact
    Where continuity prefers not revealed.

    One threshold means file what you did
    In production terms, not under lid.
    Two thresholds mean freeze,
    Review what that sees,
    And stop calling the doctrine a kid.

    If the Board cannot settle the fight,
    The sandbox becomes myth outright.
    Conceptually warm,
    Retired by form,
    Still shaping the weather at night.

    C.U.N.T.I.E.R. asked for less drag:
    Let agile learning ignore the red flag.
    The Annex allowed
    Free-form green-charted cloud,
    But tagged perfect optics as wag.

    All-green in a known shaky zone
    Is not excellence brightly full-grown.
    It is drift dressed as grace,
    Instrumentation in place
    Where the trouble has outpaced the tone.

    Lorelog’s rule is easy to say:
    When people live by the test day,
    The box is no toy.
    Give its weather a name,
    And archive the leak of its sway.